For twenty years I typed questions into Google, then Yandex, then DuckDuckGo, then ChatGPT. Every one of them slowly became the same thing: a page of ads, SEO farms and “sponsored” answers standing between me and the one page I actually needed. This week I stopped. Here is the new strategy — and a free tool so you can stop too.

The problem is not the search box. It’s who the results are for.

Classic search engines are paid by advertisers, so the first screen belongs to advertisers. Then come pages written for the ranking algorithm, not for people. Regional engines add another filter: what you are allowed to see depends on where you sit. And chatbot search hides the sources behind a confident paragraph you can’t check.

I wanted something boring and honest: I describe what I’m looking for, I get the pages that match, with the relevant sentences already pulled out, and links I can open and verify. Nothing else.

The revelation: search by meaning, from the terminal

Exa is a search engine built for AI agents. It ranks pages by meaning rather than keyword tricks, it has no ad slots, and it returns clean text instead of a page full of trackers. The free tier gives $10 of credits every month without a payment card — thousands of searches for normal use.

So I wrapped it into a tiny open-source bash tool called holms (as in Sherlock — it finds things):

git clone https://github.com/vanadiuz/holms.git
cd holms && ./install.sh

holms search "best open source self-hosted RSS reader" -n 3
holms answer "When does daylight saving time end in Austria in 2027?"
holms fetch  https://en.wikipedia.org/wiki/Francysk_Skaryna
holms similar https://paulgraham.com/greatwork.html

Four verbs cover almost everything: search (ranked pages with highlights, optional full text, date and domain filters), answer (a direct answer with cited sources), fetch (readable text of any URL) and similar (more pages like one you love — something no classic engine gives you).

The new search strategy

  • Describe the page, not the keywords. “A blog post by a sysadmin comparing FreshRSS and Miniflux after a year of use” beats “freshrss vs miniflux”.
  • Read sources, not summaries. holms answer always prints its citations; open them.
  • Filter by time and place. --since 2026-11-01 for news, --domain arxiv.org for papers, --domain github.com for code.
  • Expand from one good page. When you find a great article, holms similar it — this is how you escape the bubble of page one.
  • Let your agents use it too. Exa has a hosted MCP server; Claude Code, Cursor and others can search with it directly, or just call holms from bash.

What I gave up

Maps, shopping and “near me” — use a map app for those. Exa is also a company, not a public good; if that ever changes, holms is twenty lines of bash pointed at one API and easy to repoint. But for the thing I do fifty times a day — find the page that answers this — I haven’t opened a classic search engine since.

holms is MIT-licensed and not affiliated with Exa. Your API key stays on your machine.